testing

name	testing
description	Use when: running targeted or full tests in any workspace, choosing the safest available test command, summarizing test results, or deciding when to hand failing tests to debugger.
version	1.5
license	MIT

Testing

Skill metadata: version "1.5"; tags [testing, validation, mcp, workflow, generic]; recommended tools [testing_show_key_commands, testing_show_capabilities, testing_list_tests, testing_run_tests, testing_parse_coverage, run_in_terminal].

Procedural skill for running tests through the workspace's declared testing apparatus before, during, or after code changes.

When to use

• When a task needs targeted tests, full tests, or a validation summary
• When the workspace's test command should be discovered instead of assumed
• When deciding whether a failing test needs debugger handoff

When NOT to use

• When the user asks for a full CI-equivalent pre-commit or pre-push gate — prefer ciPreflight
• When tests are already failing and root-cause diagnosis is the main task — prefer debugger
• When the task is test coverage adequacy or regression-risk review — prefer review
• When strict Red-Green-Refactor coaching is needed and the TDD pack is installed — prefer the TDD pack skill or prompt
• When the workspace has no declared test apparatus and the task is to define one — prefer direct repo maintenance or documentation updates over this runtime workflow

Steps

1. Discover the workspace test apparatus.

   • Prefer testing_show_key_commands from the workspaceTesting MCP server.
   • Confirm a declared Run tests command exists and is not (not detected).
   • When available, call testing_show_capabilities to learn the detected runner, timeout policy, typed-targeting mode, and coverage support before choosing flags or scope.
   • If workspaceTesting is unavailable, read the workspace instructions or project docs for the documented test command; do not invent a shell command.

2. Choose the narrowest useful scope.

   • If testing_show_capabilities reports supportsTestDiscovery=true, prefer testing_list_tests to confirm available test ids before building a targeted run.
   • For a localized code change, run targeted tests first with safe argv-style target arguments only when supportsTypedTargets=true; recognized runner families may still report false for compound command shapes such as python -m unittest discover ....
   • For shared infrastructure, generated surfaces, or task completion, run the full declared test command.
   • If supportsTypedTargets=false or target arguments are uncertain or framework-specific, prefer a full run over guessing unsafe flags.

3. Execute through the safest available runner.

   • Prefer testing_list_tests(targetFiles=[...]) only when capability discovery explicitly reports safe test discovery.
   • Prefer testing_run_tests(scope="default", extraArgs=[...]) for targeted runs.
   • Treat targetFiles and testNames as valid only when supportsTypedTargets=true; otherwise expect workspaceTesting to reject them and use scope="full" or explicit extraArgs instead.
   • Prefer testing_run_tests(scope="full") for full runs.
   • Use the documented native command only when the workspaceTesting server is unavailable.
   • Do not use arbitrary command strings, shell interpolation, pipes, redirects, or commands requiring secrets unless the workspace documentation explicitly requires them.

4. Summarize the result in an agent-readable form.

   • Report status, command or tool used, exit code, runnerExitReason when present, and the relevant stdout/stderr tail.
   • Identify first failing test, failing file, or failure class when the output makes that clear.
   • State whether the check is targeted, full, skipped, unavailable, or blocked.

5. Parse coverage artifacts when they are part of the workspace's test apparatus.

   • Prefer testing_parse_coverage(coveragePath="coverage.xml") for Cobertura-style coverage output when the file exists or is documented by the workspace.
   • Report percent covered, valid/covered line counts, and zero-coverage files when available.
   • If no coverage artifact is declared or produced, state that coverage was not assessed rather than guessing.

6. Act on failures deliberately.

   • If the failure is clearly caused by the current edit slice, fix that slice and rerun the same test.
   • If root cause is unclear, delegate to debugger with the exact command, output tail, changed files, and why the test was selected.
   • If a test cannot run because the workspace lacks a declared test command, report the missing apparatus and suggest adding one to workspace instructions.

Verify

• A declared workspace test command or explicit unavailable state has been observed this session
• testing_show_capabilities was used or an explicit reason was given for skipping capability discovery
• testing_list_tests was used only when supportsTestDiscovery=true, or its absence was justified by capability output
• workspaceTesting.testing_run_tests was preferred when available, with documented native fallback only when needed
• Targeted runs used safe argv-style arguments; full runs used the command exactly
• Output summary includes status, command/tool, exit code when available, runnerExitReason when present, and relevant output tail
• Coverage artifact was parsed with testing_parse_coverage when present or explicitly reported as not assessed
• Failing tests were either repaired in the current slice or handed to debugger with exact evidence